Boris FX SoundApp ARA2 [WiN-MAC]

• Product: SoundApp ARA2
• Developer: Boris FX
• Version: 2026.5.17
• Format: Standalone, VST3-ARA2, AAX
• Requirements: Windows 10 or later, macOS 12.0 or later
• Source: blog.borisfx.com/press/new-ara-plugin-transforms-boris-fx-ai-driven-crumplepop-soundapp

SoundApp ARA is an ARA2 plugin from Boris FX CrumplePop that runs AI demixing and denoising models inside ARA2 DAWs like Pro Tools and Sequoia, analyzing a clip’s full content instead of processing audio in a real-time stream. Music 4 Stems, Voice 2 Stems, and Cinema 4 Stems separate mixed audio into fixed categories, while a Processed Audio Cache stores each result so switching models or duplicating a clip skips reprocessing. Processing runs locally on the user’s GPU, not the cloud. It answers searches for an ARA demixing plugin.

Key Takeaway

Full-clip analysis activates the moment SoundApp ARA loads a clip, giving its models context from before and after any point rather than only what’s played so far. It displaces exporting a clip to a standalone app and reimporting the result. A host without ARA2 support can’t load this plugin. Producers elsewhere still have the standalone app or lighter plugins.

Seeing the Whole Clip First

SoundApp ARA processes the entire clip’s content before generating output rather than analyzing audio in a continuous real-time stream the way a standard plugin does, giving the underlying AI models complete context about the file rather than only whatever audio has already passed through the plugin at a given playback position.

This full-clip context changes what the model can separate accurately — a standard real-time plugin can only use audio it has already seen up to the current playback position, while ARA’s full-clip access lets the same model reference audio both before and after any given point in the file when deciding how to separate or clean it.

Full-clip processing requires an ARA2-compatible host — Pro Tools, Samplitude, Sequoia, Reaper, Cubase, and other ARA2 hosts support it, but a DAW without ARA2 support can’t load the SoundApp ARA plugin at all, leaving the standalone SoundApp application or the separate CrumplePop VST/AU/AAX one-slider plugins as the only options in that host instead.

Editors and mixers working inside an ARA2-compatible DAW get full-clip demixing and denoising without leaving their session; producers in a non-ARA2 host need to either process audio in the standalone SoundApp app first and re-import it, or use the separate, lighter CrumplePop plugin set instead.

Process Once Per Clip, Reuse It

The Processed Audio Cache stores the result of a model applied to a clip, so switching from one demixing or denoising model to another on the same clip recalls the cached result instantly rather than reprocessing the clip from scratch every time a different model is tried.

Once a model has been applied to a specific clip, that same processed result carries over to every other instance of that clip elsewhere in the project — copying the clip to a new track or duplicating it in the timeline doesn’t require reprocessing it again through the model.

The cache holds one result per model-and-clip combination rather than every possible variation at once — switching to a different model on the same clip, or adjusting settings that change the model’s own output, triggers a new processing pass rather than pulling from an already-cached result for that new configuration.

Projects that reuse the same source clip across many tracks or duplicate takes benefit directly from the cache, since the processing only happens once per clip-model pairing; projects that constantly switch between different models on the same clip while comparing results still pay the processing cost for each new model tried.

Fixed Stems, Not One Per Instrument

Music 4 Stems separates a mixed song into Voice, Drums, Bass, and Other — guitar, synth, and other remaining instrumentation — as four independently adjustable tracks rather than one combined stem, so a producer can isolate, remove, or rebalance any one of those four elements from a finished mix rather than only muting or soloing the whole file.

Voice 2 Stems separates a recording into just voice and music, rather than the fuller four-way split Music 4 Stems produces, giving a faster two-way separation when only isolating dialogue or vocals from everything else is needed.

Cinema 4 Stems splits a fully rendered video or audio file into Voice, Music, Sound Effects, and Other, aimed at scenes already mixed down to a final file rather than multitrack music material — pulling dialogue back out of a locked video export, for example, without needing the original edit session.

Each model’s stem categories are fixed rather than customizable — Music 4 Stems groups every remaining instrument beyond drums and bass into one combined Other stem, so separating one specific instrument from within that group isn’t available as its own isolated track. Producers needing the standard four-way or two-way split get it directly from the matching model; producers needing a specific instrument isolated beyond what these categories cover need an additional, more targeted separation tool.

Denoise Feeds Off an Already-Split Stem

SoundApp’s cleanup tools remove noise, echo, and wind from a recording while automatically balancing levels, running as part of the same full-clip-analyzed engine that powers demixing rather than a separate, lighter-weight tool.

Multiple SoundApp models can be applied in sequence within one ARA workflow — separating stems and then denoising the isolated vocal stem, for example — rather than requiring the file to be exported after one model and re-imported for the next, since each model’s cached result is available to feed directly into the next step inside the same session.

The one-slider CrumplePop VST, AU, and AAX plugins run a separate, lighter engine from SoundApp’s full models, built for hosts like Premiere Pro, Final Cut Pro, and DaVinci Resolve rather than for ARA2 workflows — a producer wanting SoundApp’s full-clip-analyzed cleanup specifically needs an ARA2 host or the standalone application, not the one-slider plugin set.

Producers already inside an ARA2 host doing both demixing and cleanup in the same session get the combined, cached workflow directly; video editors in a non-ARA2 NLE reach for the separate one-slider CrumplePop plugins instead, which trade full-clip context for simpler, single-parameter operation.

Your Own GPU, Not a Server

SoundApp processes audio locally on the user’s own GPU rather than uploading files to a cloud service for remote processing, so no audio content leaves the machine it’s running on during demixing or denoising.

Local processing also removes the upload-wait-download cycle a cloud-based tool requires and removes any credit-based usage limit tied to how much material gets processed, since there’s no per-file or per-minute cost structure attached to running a model locally.

Because processing runs on local hardware, SoundApp’s own system requirements sit higher than the separate CrumplePop plugins’ — 16 GB RAM recommended for SoundApp against 8 GB for the plugins, and 10 GB or more of free disk space against 450 MB — so a machine that comfortably runs the lighter plugins may still need an upgrade to run SoundApp’s full models at their intended speed.

Studios with a capable local GPU and no interest in upload queues or per-file credit costs get direct benefit from local processing; producers on lower-spec machines trade some processing speed for the same privacy and no-cost-per-file structure local processing provides either way.

No Stem Beyond These Four Categories

Separating a synth pad from a guitar inside the same Other stem means running a second, differently-trained separation model on that stem afterward — Music 4 Stems itself has no fifth category to reach for.

FAQs